Germination Timing Influences Natural Selection on Life-History Characters in Arabidopsis Thaliana

Ecology, 83(4), 2002, pp. 1006±1016 ᭧ 2002 by the Ecological Society of America

GERMINATION TIMING INFLUENCES NATURAL SELECTION ON LIFE-HISTORY CHARACTERS IN ARABIDOPSIS THALIANA

KATHLEEN DONOHUE1 T. H. Morgan School of Biological Sciences, 101 Morgan Building, University of Kentucky, Lexington, Kentucky 40506 USA

Abstract. An experimental manipulation of germination timing was conducted to test whether germination timing in¯uences the phenotypic expression of postgermination life- history characteristics and whether it alters natural selection on those characters. Seeds collected from ®ve natural populations of Arabidopsis thaliana in Kentucky were forced to germinate in early November, early December, and early March. Life-history characters such as timing of reproduction, size at reproduction, and size at senescence were measured, and fruit production and mortality were monitored. Germination timing signi®cantly altered subsequent life-history characters and reproduction. November germinants were larger than December germinants when they began reproducing, and they commenced reproduction sooner. All spring germinants died before reproducing. Germination timing also in¯uenced natural selection on life-history characters. December germinants were more strongly selected to be large at the time of reproduction than November germinants, indicating stronger selection for a faster growth rate in December germinants. Stabilizing selection on timing of reproduction was detected in December germinants but not in November germinants. Therefore, variation in germination timing in¯uenced ®tness, modi®ed the phenotypic expression of important life-history characters, and altered the strength and mode of natural selection on them. Key words: Arabidopsis thaliana; dormancy; germination; habitat selection; life-history characters; maternal characters; phenology; phenotypic plasticity; seasonal dormancy; variable selection.

INTRODUCTION Venable 1984, Kalisz 1986, Kalisz and Wardle 1994, Timing of germination is highly responsive to en- Nordborg and Bergelson 1999). Variation in the timing vironmental conditions. Consequently, if environments of germination and ¯owering may even lead to varia- change, due to habitat alteration or global warming for tion in the number of generations per year. Thus de- instance, germination behavior is likely to change as a termining the effects of variation in germination timing direct and immediate response. For example, in species is useful for explaining the persistence of variation in that typically germinate in the early autumn, protracted other life-history characters. summer drought conditions could cause germination to Many plants, including many important weeds, typ- be delayed until later in the autumn. Such a delay may ically display a ``winter annual'' life history, in which not only have direct ®tness consequences, but it may seeds germinate in the autumn, small plants overwinter also affect other aspects of the plants' life history. Ger- as rosettes, and in the springtime they grow, ¯ower, set mination timing may also be altered by evolutionary seeds, and die as the summer approaches. Other plants responses to selection on germination. Characterizing display a ``summer annual'' life history, in which seeds how germination timing in¯uences ®tness, phenotypic germinate in the early spring, and they ¯ower, set seeds, expression, and natural selection on life-history char- and die all during the same season. This basic life- acters therefore provides information on the manner in history sequence varies among known A. thaliana eco- which plants might be affected by environmentally in- types worldwide (Ratcliffe 1965, Effmertova 1967, duced or evolutionary changes in their germination be- Evans and Ratcliffe 1972, Nordborg and Bergelson havior. 1999), as well as among natural populations in North Plants vary greatly in their life histories, including America (L. Dorn and K. Donohue, personal obser- the size and timing of reproduction, the number of re- vations). Populations in Kentucky have been charac- productive bouts, and the number of generations com- terized as winter annuals that germinate in autumn pleted in a single growing season. Germination timing (Baskin and Baskin 1983), whereas New England is closely associated with subsequent life-history char- (Rhode Island) populations have been seen to germi- acters in plants, such as annual vs. biennial strategies nate either in autumn or in spring (L. Dorn and K. and summer vs. winter annual habits (Chouard 1960, Donohue, personal observation). The mechanism of ``summer annual'' vs. ``winter annual'' life histories in Manuscript received 27 November 2000; revised 12 June 2001; accepted 4 July 2001. A. thaliana is likely to be a function of both seed dor- 1 E-mail: [email protected] mancy and vernalization requirements for ¯owering 1006 April 2002 GERMINATION TIMING INFLUENCES SELECTION 1007

(Napp-Zinn 1976, Nordborg and Bergelson 1999). ``TC,'' was collected from a fallow ®eld which had not Therefore, investigating the relationship between the been cultivated for approximately two years. One pop- season of germination and subsequent life-history char- ulation, ``RR,'' was collected from an older fallow ®eld acters is relevant for understanding the basic ecology with substantial secondary succession and a vegetation of summer and winter annual strategies. canopy of herbaceous annuals and perennials. The last Seasonal seed dormancy can delay germination until population, ``Garden,'' was collected from plants grow- conditions are appropriate for growth and consequently ing between bricks in a garden. Seeds were collected may strongly in¯uence ®tness. In temperate climates, in May 1999 and were stored dry at room temperature delaying germination until spring can prevent over- until they were planted. winter mortality. Autumn germination, however, may Seeds were planted into 96-well plug trays ®lled with provide a selective advantage when the risk of winter Strong-Lite bedding plant mix (Strong-Lite Horticul- mortality is low by enabling the plant to ¯ower earlier tural Products, Seneca, Illinois, USA). All seeds were in the spring or at a larger size. Consequently, selection forced to germinate synchronously by stratifying seeds for spring vs. autumn germination depends on the rel- on soil for three days at 5ЊC. After cold strati®cation, ative risks of winter mortality and the selective advan- seeds were placed in a growth chamber at 25ЊC with tage of being developmentally advanced in the spring full-spectrum light on a 12-h photoperiod. After ger- (Masuda and Washitani 1992). Germination timing mination, seeds were transferred to an unregulated within a season, moreover, may in¯uence ®tness by greenhouse for three days for acclimation and then altering the size of overwintering and the size attained transferred to a ®eld plot on the University of Ken- at the beginning of the spring growing season. Mea- tucky's Lexington campus. Seeds from the ®ve popu- suring the ®tness consequences of the seasonal timing lations were randomly distributed within three plug of germination can elucidate the selective mechanisms trays (blocks) per treatment, with 18 replicates per pop- for the prevalence of these particular life-history strat- ulation per block in each treatment. This gave a total egies. of 810 seedlings. All plug trays were kept at a moisture Germination timing not only can in¯uence ®tness level comparable to that of the surrounding vegetation directly, but it can in¯uence the selective environment throughout the season through occasional supplemental experienced by the plant after germination. If seeds watering since the plugs tended to dry out slightly more germinate only under particular conditions, then the quickly than the surrounding vegetation. seeds effectively choose the selective environment that As a phenotypic manipulation of germination timing, determines the evolution of postgermination traits. In three germination treatments were imposed: early au- this manner, genetically based associations can evolve tumn, late autumn, and spring. ``Early autumn'' ger- between germination and postgermination characters minants were forced to germinate (placed at 5ЊC) in (Evans and Cabin 1995). Germination timing therefore early November and were transferred into the ®eld on is likely to in¯uence the evolution of life-history char- 14 November. Natural germination occurs primarily in acters in plant populations by in¯uencing the expres- mid-October in Kentucky (Baskin and Baskin 1983), sion of life history variation and by determining the so the early autumn germination treatment was ap- selective environment that acts on that life-history var- proximately two and a half weeks behind the peak ger- iation. mination time of the natural populations in the ®eld. Using ®ve populations of Arabidopsis thaliana col- However, germination in the ®eld continues at least lected from central Kentucky, I conducted an experi- through the third week of November (K. Donohue, per- mental manipulation of germination timing to assess sonal observation), so these experimental seedlings its effects on life-history characters and ®tness. Spe- germinated well within the natural range of germina- ci®cally, I asked the following questions. Does ger- tion timing. This treatment re¯ects the most natural mination timing in¯uence the phenotypic expression of germination timing in this experiment. ``Late autumn'' life-history characters? Does germination timing in¯u- germinants were forced to germinate during late No- ence mortality and reproduction? Does germination vember and were transferred into the ®eld on 4 De- timing alter natural selection on postgermination char- cember. This treatment delayed germination beyond the acters? germination timing naturally observed in the ®eld and thereby functioned to extend the range of variation to METHODS include phenotypes that may have already been selectively eliminated from the population (Wade and Kalisz Experimental design 1990). This treatment simultaneously manipulated both Seeds were collected from ®ve natural populations the timing of germination and the size of overwinter- of Arabidopsis thaliana around Lexington, Kentucky, ing, since later germinants are necessarily smaller at USA. Two populations, ``Hort'' and ``Ag,'' were col- the onset of winter conditions. Experimentally extend- lected from agricultural ®eld edges at the University ing the range of variation in these two traits permitted of Kentucky's Horticultural Research Station and Ag- more powerful tests of whether germination timing and ricultural Research Farm, respectively. One population, size of overwintering in¯uences mortality over the win- 1008 KATHLEEN DONOHUE Ecology, Vol. 83, No. 4 ter or growth and reproduction in the spring. Spring differed due to either genetic differences between pop- germinating seeds were cold treated in late February ulations or to maternal effects of ®eld-collected seeds. and were transferred into the ®eld on 8 March. This A population by treatment interaction indicates that treatment represents the germination timing observed populations differ in their response to germination by individuals in some New England populations with treatment. All results reported are based on Type III summer annual strategies. sums of squares. Analysis of fruit number included all The diameter of each seedling was recorded on the plants, including those with no fruits. The spring ger- day that it was transferred to the ®eld. The number of minants did not naturally express any of the phenotypes leaves was counted and the rosette diameter was mea- because they never bolted. Therefore, only the early sured on the early autumn germinants approximately and late autumn germinants were compared in the anal- three weeks later, when the late autumn germinants yses presented in the tables. were placed into the ®eld. The number of rosette leaves A Kaplan-Meier survival analysis using Proc Lifetest was recorded for all autumn germinants ®ve weeks later (SAS 1990) was used to compare mortality schedules as an estimate of overwintering size. The number of across germination treatments. The data were uncen- rosette leaves and rosette diameter were recorded on sored. all plants when the spring germinants were transferred A phenotypic selection analysis (Lande and Arnold into the ®eld. The following characters were measured 1983) was performed to determine the magnitude of on all plants that expressed them: bolting date (the date natural selection on the measured characters. Rosette on which the apical meristem begins to develop into diameter, bolting date, the time interval between ¯ow- the in¯orescence, which signi®es a switch from veg- ering and bolting, and total number of branches were etative to reproductive allocation of the meristem), used in the analysis. Two characters were excluded number of rosette leaves at bolting, rosette diameter at from the selection analysis a priori: The number of bolting, ¯owering date, and date of death. The follow- rosette leaves was not used because, like rosette di- ing characters were determined at senescence: number ameter, it is a measure of size at the time of reproduc- of basal branches (branches originating from the ro- tion. Flowering time also was not used because of co- sette), in¯orescence branches (branches on the main linearity with bolting date and ¯owering interval. Char- in¯orescence stem), secondary and higher level branch- acters were standardized to have a mean of zero and es, and total number of fruits, as an estimate of total standard deviation of one within each treatment. Rel- lifetime ®tness. ative ®tness was calculated within each treatment by The spring germinants did not bolt by the time that dividing the total number of fruits produced by an in- most of the plants had died. One block of the spring dividual by the mean number of fruits of all individuals germinants was kept alive by watering in order to de- within a given treatment. If the individual expressed termine the phenotype they would express if the dry the phenotypes but did not set fruit, it had a ®tness of summer conditions were postponed. These plants were zero. Population was included as a ®xed effect in the monitored until they died, and the same characters as analysis in order to control for differences among pop- mentioned above were recorded. ulations in unmeasured characters that in¯uence ®tness (Donohue et al. 2000). Direct selection was estimated Statistical analysis by analysis of covariance that included all traits, with The SAS (SAS Institute 1990) statistical package relative ®tness as the dependent variable. Total selec- was used for all analyses. Characters were transformed tion (direct selection plus indirect selection through to normality when necessary. A multivariate analysis correlations with other characters) was estimated in a of covariance was used to determine whether the mea- model that included only one trait at a time. In addi- sured characters differed among populations and ger- tional analyses, selection coef®cients were compared mination treatments. Separate analyses of covariance between early and late autumn germination treatments then determined which of the traits were signi®cantly using analysis of covariance in which germination in¯uenced by these factors. Block was nested within treatment was included as a ®xed factor; signi®cant treatment, giving a split-plot design. Population and differences between treatments in selection coef®cients germination treatment were considered ®xed effects, were apparent as signi®cant interactions between the and block was considered a random effect. Main effects character and germination treatment (Donohue et al. of germination treatment were tested over the block 2000). These estimates of the strength of direct and and error mean squares, and main effects of population total selection are not completely comparable to those were tested over block±population interactions and er- of Lande and Arnold (1983). They differ from them ror mean squares. Despite efforts to transfer seedlings due to the inclusion of population as a main effect in into the ®eld at the same size at the different planting the analyses and the use of regression coef®cients from times, there was signi®cant variation in initial size analysis of covariance, rather than covariance, for an among the planting treatments. Therefore, initial seed- estimate of total selection. This method, however, has ling size was used as a covariate in these analyses. A the advantage of controlling for effects of unmeasured population main effect indicates that the populations characters that differ between populations that may si- April 2002 GERMINATION TIMING INFLUENCES SELECTION 1009

though the difference between treatments decreased over time, as indicated by signi®cant interactions between germination treatment and census (rosette diameter, F ϭ 582.41, df ϭ 2, 512, P Ͻ 0.001; number of rosette leaves, F ϭ 7.78, df ϭ 2, 512, P Ͻ 0.01). Both early and late autumn germinants were substan- tially larger than the spring germinants when spring germinants were put into the ®eld. Populations differed in the number of rosette leaves throughout the winter. Populations also differed in the degree to which the number of rosette leaves was reduced in the late autumn germination treatment, as indicated by a signi®cant population by germination treatment interaction. Pop- ulations ceased to respond differently to the autumn germination treatments by the time the spring germinants were put into the ®eld, however. Growth trajec- tories throughout the winter and spring also differed among populations, as indicated by the signi®cant interaction between population and census in a repeated measures analysis of leaf number. Such differences were expressed only in early autumn germinants (F ϭ 3.33, df ϭ 8, 504, P ϭ 0.001; based on Wilk's ␭) but not in late autumn germinants (F ϭ 1.10, df ϭ 8, 504, P Ͼ 0.05; based on Wilk's ␭). Thus, experimentally determined timing of germination resulted in persistent size differences among the treatments throughout the winter and into spring, but early in the experiment the magnitude of the difference in size depended on the population the seedling was from. FIG. 1. Size throughout the winter and early spring. Ro- sette diameter at the time at which the seedlings were placed Treatment effects on postgermination life-history into the ®eld is shown in the upper ®gure. The number of rosette leaves throughout the winter and spring for each of characters and ®tness the treatments is shown in the lower ®gure. For all treatments, Experimentally determined germination timing sig- the seedlings had no rosette leaves when they were placed into the ®eld; hence the bar is not visible for those treatments ni®cantly altered the expression of postgermination with no rosette leaves. T refers to a main effect of germination life-history characters (Table 1, Fig. 2). MANOVA treatment; P refers to a main effect of population; P ϫ T detected signi®cant effects of autumn germination refers to a population-by-treatment interaction. treatment (F ϭ 73.27, df ϭ 1, 500, P Ͻ 0.001) and * P Ͻ 0.05; ** P Ͻ 0.01; *** P Ͻ 0.001. population (F ϭ 21.33, df ϭ 4, 500, P Ͻ 0.001) on life-history characters. Early autumn germinants had more rosette leaves than late autumn germinants at the time of reproduction, and the switch from vege- multaneously in¯uence both phenotype and ®tness, and tative to reproductive growth (bolting time) was ear- inclusion of ®xed treatment factors in the additional lier in early germinants. Floral development time, or analyses enables direct tests for signi®cant differences the interval between bolting and ¯owering, was not in selection in different treatments. Selection analysis signi®cantly in¯uenced by germination timing, so ear- was performed separately on the spring germinants that ly autumn germinants also ¯owered signi®cantly ear- were watered because the unwatered spring germinants lier than late autumn germinants. Within the treat- did not express the phenotypes. ments, the initial size at which the seedling was put Pearson correlations were calculated for all traits into the ®eld in¯uenced the size attained at reproduc- within each germination treatment. The spring germi- tion and the timing of reproduction in the spring. nants that were analyzed were those that had been kept Fitness was in¯uenced by the timing of germina- alive through supplemental watering. tion. No early autumn germinants exhibited any sign RESULTS of frost damage. Some frost damage was apparent in the late autumn germinants; meristems appeared to be Starting conditions and characterization damaged in 1.9% of the plants. Overwinter mortality of the treatments was low in both germination classes. Early autumn Early autumn germinants remained larger than late germinants had Ͻ1.0% overwinter mortality, and autumn germinants throughout the winter (Fig. 1), al- 2.7% of the late autumn germinants died during the 1010 KATHLEEN DONOHUE Ecology, Vol. 83, No. 4

TABLE 1. Results of analysis of covariance to test for effects of germination timing and population on morphological and phenological characters. Only early and late autumn germination treatments are included.

Germination Germination time ϫ Trait Initial size Block time Population Population No. rosette leaves 61.42*** 3.27* 12.57** 12.02*** 0.63 Rosette diameter 0.71 1.23 2.86 1.70 0.25 No. branches 0.25 5.33** 0.01 5.07** 1.23 Bolting date 28.42*** 13.71** 99.38*** 125.73*** 0.92 Flowering interval 7.76** 17.44*** 1.97 8.22*** 6.80** Flowering date 19.71*** 26.77*** 68.69*** 194.84*** 0.95 No. fruits 2.37 1.71 3.47² 0.51 0.82 Notes: The F ratios given are based on Type III sums of squares. Numerator df ϭ 1 for initial size and germination time, df ϭ 2 for block, df ϭ 4 for population and germination time ϫ population. Denominator df ϭ 503 for all factors except germination time. Denominator df for germination time ranged from 5 to 25 due to missing values. ² P ϭ 0.08; * P Ͻ 0.05; ** P Ͻ 0.01; *** P Ͻ 0.001.

FIG. 2. Means (Ϯ 1 SE) of life-history traits of early autumn germinants, late autumn germinants, and spring germinants that had been kept alive through supplemental watering. Each line connects the mean value of a population in the three treatments. Populations are described in Methods: Experimental design. Characters related to the size at reproduction are shown in panels A±C. Characters related to the timing of reproduction are shown in panels D±F. Rosette diameter was measured in cm; bolting date and ¯owering date are expressed as day of year. April 2002 GERMINATION TIMING INFLUENCES SELECTION 1011

FIG. 3. Mortality schedules of plants that germinated in early autumn, late autumn, and spring. Only one plant died before 1 January (not shown).

winter. Mortality schedules did not differ signi®cantly supplemental watering. The development of the wa- among autumn germination treatments (Fig. 3), and tered plants, moreover, was noticeably altered (Fig. 2). most mortality was episodic and corresponded with Eventually, 56% of these plants bolted, but some lived stressful drought conditions in the summer (P Ն 0.05 for many weeks without bolting. Bolting time was sig- in Kaplan-Meier survival analysis comparing early ni®cantly later than early and late autumn germinants and late autumn germinants). Ninety-six percent of (P ϭ 0.0001 for both comparisons). Plants that did bolt both early and late autumn germinants survived to had many more rosette leaves and had larger diameters bolt. Early autumn germinants produced a mean of 12 by the time they bolted than did early and late autumn more fruits (ϳ240 more seeds) than late autumn ger- germinants (P ϭ 0.0001 for both comparisons). They minants, but this difference was only nearly signi®- also had signi®cantly more branches (P ϭ 0.0001 for cant (Table 1, Fig. 4). both comparisons). Fruit production of watered spring The spring germinants died slightly earlier than the germinants was much less than that of either of the autumn germinants (Fig. 3), as indicated by a signi®- autumn germinants (Fig. 4; P ϭ 0.0001 for both com- cant effect of germination timing in a Kaplan-Meier parisons). survival analysis (P ϭ 0.01 in an analysis that included Populations differed signi®cantly in all life-history all three treatments, excepting the spring germinants characters except rosette diameter (Table 1, Fig. 2). In that had supplemental water). Every spring germinant particular, plants from the population collected from died before switching from vegetative growth to re- the bricks in a garden bolted and ¯owered signi®cantly production, except for those that were kept alive by earlier and at a smaller size than did plants from the

FIG. 4. Means and standard errors of fruit production by early autumn germinants, late autumn germinants, and spring germinants that had been kept alive through supplemental watering. All unwatered spring germinants had no fruit production. Each line connects the mean value of a population in the three treatments. 1012 KATHLEEN DONOHUE Ecology, Vol. 83, No. 4

TABLE 2. Results of phenotypic selection analysis.

Early autumn Late autumn Germination Germination Total Direct Total Direct timing ϫ total timing ϫ direct selection selection selection selection selection selection

Character (S) (␤) (S) (␤) (FS)³ (F␤)§ Rosette diameter 0.010*** 0.007*** 0.023*** 0.018*** 32.22*** 22.19*** No. branches 0.012*** 0.010*** 0.023*** 0.013*** 21.08*** 1.77 Bolting date 0.004² 0.001 0.000 0.001 0.02 0.29 Flowering interval 0.003² 0.004** 0.004 Ϫ0.001 0.17 2.90ϩ Notes: Selection differentials measuring the magnitude of total selection (S) and selection gradients measuring the magnitude of direct selection (␤) are given. F ratios are given for the tests for signi®cant differences between germination timing treatments in the magnitude of total selection (FS) and direct selection (F␤). Signi®cant nonlinear selection coef®cients for univariate (g) and multivariate (␥) selection analyses are as follows. Number of branches for early germinants (g ϭ 0.007, P Ͻ 0.001; ␥ϭ0.005, P Ͻ 0.001), number of branches for late germinants (g ϭ 0.008, P Ͻ 0.001; ␥ϭ0.005, P Ͻ 0.001), bolting time for late germinants (g ϭϪ0.006, P Ͻ 0.001). N ϭ 258 for early autumn germinants. N ϭ 260 for late autumn germinants. ² P ϭ 0.1; * P Ͻ 0.05; ** P Ͻ 0.01; *** P Ͻ 0.001. ³dfϭ 1, 511; § df ϭ 1, 502. other populations. Populations did not differ in total since rosette diameter was also under strong positive fruit production. Populations responded differently to direct selection. Signi®cant disruptive selection was the autumn germination treatments. Plants from the detected for branch production in both early and late garden population did not increase the interval between autumn germinants, but the phenotype with the mini- bolting and ¯owering as much as the other populations mum ®tness was not within the range of phenotypic in the early autumn treatment. In the watered spring variants in this study. germinants, signi®cant differences between popula- Signi®cant stabilizing selection was detected for tions were detected for rosette diameter (F ϭ 3.51, df bolting time in late autumn germinants, and the optimal ϭ 4, 40, P Ͻ 0.05), bolting time (F ϭ 3.85, df ϭ 4, phenotype was within the range of phenotypic variants 40, P Ͻ 0.01), and ¯owering time (F ϭ 4.37, df ϭ 4, (Table 2). Therefore an intermediate bolting time was 40, P ϭ 0.005). optimal in this sample. Early autumn germinants exhibited signi®cant direct selection (␤) and nearly sig- Natural selection on postgermination characters ni®cant total selection (S) for a shorter time interval The strength of natural selection on life-history char- between bolting and ¯owering. However, the difference acters was signi®cantly in¯uenced by the timing of in the strength of selection on this character between germination (Table 2). Plants that germinated later in early and late autumn germinants was only nearly sig- the autumn were more strongly selected to be larger ni®cant. (have a larger rosette diameter) upon switching from Some relationships among life-history characters vegetative growth to reproduction. There was also differed between early and late autumn germinants (Ta- stronger total selection (S) for larger overall size in ble 3). Bolting date was signi®cantly positively cor- these plants, as indicated by the signi®cantly stronger related with the size at which reproduction was initiated total selection for more branches and larger rosettes. (number of rosette leaves and rosette diameter) in early The stronger total selection for more branches in late autumn germinants but not in late autumn germinants, autumn germinants could be due in part to the some- suggesting that late autumn germinants that delayed what stronger positive correlation between rosette di- bolting did not necessarily attain a large size. In con- ameter and branch production in these plants (Table 3), trast, the number of leaves at bolting signi®cantly in-

TABLE 3. Pearson correlations among characters for early autumn germinants (below diagonal) and late autumn germinants (above diagonal).

Bolting Flowering Character No. leaves Diameter No. branches date Interval date No. fruit No. leaves 1.00 0.43*** 0.34*** Ϫ0.01 0.12 0.05 0.31*** Diameter 0.20** 1.00 0.49*** Ϫ0.04 0.08 Ϫ0.01 0.60*** No. branches Ϫ0.07 0.31*** 1.00 0.09 0.19** 0.19** 0.59*** Bolting date 0.25*** 0.30*** 0.19** 1.00 Ϫ0.33*** 0.95*** 0.05 Flowering interval 0.21*** Ϫ0.04 0.02 Ϫ0.20*** 1.00 Ϫ0.01 0.09 Flowering date 0.35*** 0.26*** 0.19** 0.85*** 0.34*** 1.00 0.05 No. fruit 0.05 0.45*** 0.56*** 0.08 0.11 0.14* 1.00 Notes: Boldface indicates signi®cance after Bonferroni corrections for multiple comparisons. N ϭ 258 for early autumn germinants. N ϭ 260 for late autumn germinants. * P Ͻ 0.05; ** P Ͻ 0.01; *** P Ͻ 0.001. April 2002 GERMINATION TIMING INFLUENCES SELECTION 1013

TABLE 4. Results of phenotypic selection analysis of spring germinants that had been watered to prolong their growing season.

Nonlinear selection coef®cients Selection Selection Univariate Multivariate Character differential (S) gradient (␤) (g) (␥) Rosette diameter Ϫ0.06 0.01 0.02 0.02 No. branches 0.23*** 0.30*** 0.09** 0.09* Bolting date Ϫ0.05 Ϫ0.04 Ϫ0.06 0.05 Flowering interval Ϫ0.07 Ϫ0.19 Ϫ0.07 Ϫ0.40 Notes: Selection differentials (S) and selection gradients (␤) are given. Nonlinear selection coef®cients are given for univariate (g) and multivariate (␥) selection analyses; N ϭ 50. * P Ͻ 0.05; ** P Ͻ 0.01; *** P Ͻ 0.001.

¯uenced the total number of reproductive branches and DISCUSSION ®tness in late autumn germinants but not in early autumn germinants. It therefore appears that the variation Life-history characters and ®tness were shown to in the size attained before reproduction by the late au- depend on the timing of germination. A difference in tumn germinants more strongly in¯uenced their overall germination timing of three weeks in the autumn sig- size and reproductive output, whereas reproduction at ni®cantly altered basic life-history characters such as a consistently more advanced stage by early autumn the timing of reproduction and size at reproduction. germinants mitigated the association between size at Germination timing also clearly had strong ®tness con- reproduction and size at senescence. sequences, with the early autumn germinants having A different pattern of selection was found for spring the highest ®tness. Even though late autumn germinants germinants that were kept alive beyond the natural were signi®cantly smaller than early autumn germi- growing season (Table 4). In these plants, only the total nants throughout the winter, overwinter mortality dif- number of reproductive branches was associated with fered very slightly between early and late autumn ger- ®tness. Signi®cant disruptive selection was detected for minants. The size that seedlings attained for overwin- branch production, but the phenotype with minimum tering therefore did not determine the degree of over- ®tness was outside of the range of phenotypic variants winter mortality. Instead, the marginally nonsigni®cant in this study. Size attained at reproduction and the tim- difference in ®tness between early and late autumn ger- ing of reproduction had no signi®cant in¯uence on ®t- minants was due to the difference in size attained be- ness in this sample that did not experience summer fore reproduction, which then in¯uenced total fruit pro- drought conditions. Correlations among characters duction. were quite weak in general (Table 5), indicating atyp- Much of the in¯uence of germination timing on life- ical variation in the characters and re¯ecting the ap- history expression appears to be mediated by size. Al- parently disrupted developmental pattern. These results though the differences in size between germination indicate that natural selection on the measured char- treatments decreased over time, size differences be- acters was relaxed when the growing season was ar- tween treatments persisted throughout the life history ti®cially prolonged or that life-history characters that of the plants in this experiment. Later germinants ap- were selectively important under more typical condi- pear to have grown slightly faster than early germi- tions contributed little to ®tness variation when normal nants, reducing size differences in the spring, but this development was disrupted. difference in growth rate was not suf®cient to eliminate

TABLE 5. Pearson correlations among characters for spring germinants that had been watered to prolong their growing season.

No. No. Bolting Flowering Flowering No. Character leaves Diameter branches date interval date fruits No. leaves 1.00 0.27 0.23 0.05 0.06 Ϫ0.02 0.09 Diameter 1.00 Ϫ0.07 0.16 0.27 0.15 Ϫ0.18 No. branches 1.00 0.11 0.01 0.21 0.71*** Bolting 1.00 0.31* 0.97* Ϫ0.10 Interval 1.00 0.21 Ϫ0.21 Flowering 1.00 Ϫ0.05 No. fruits 1.00 Notes: Boldface indicates signi®cance after Bonferroni corrections for multiple comparisons; N ϭ 50. * P Ͻ 0.05; ** P Ͻ 0.01; *** P Ͻ 0.001. 1014 KATHLEEN DONOHUE Ecology, Vol. 83, No. 4 size differences due to variation in germination timing short strati®cation and long vernalization duration. in the autumn. This result suggests adaptive coordination of germi- Natural selection on life-history characters also de- nation and vernalization requirements in nondisrupted pended on the timing of germination. Selection on the plants, but determining the role of germination timing size attained before switching to reproduction was in mediating selection on vernalization requirements stronger in plants germinating in the late autumn than requires further study. in those germinating earlier in the autumn, suggesting The populations differed signi®cantly in their life- stronger selection for increased growth rate in later history characters and in the plasticity of one of these germinants. In addition, stabilizing selection on bolting characters (the time interval between bolting and ¯ow- time was detected in late autumn germinants but not ering) in response to germination treatment. Population in early autumn germinants. Bolting too early at a small differences expressed in the spring germination treat- size apparently caused a decrease in size-dependent ment with a protracted growing season suggests that reproduction, whereas bolting too late caused mortality genetic variation may exist for life-history expression before fruit maturation was completed. These patterns in novel environments (such as environments with very of selection and the failure of spring germinants to short cold spells and long growing seasons). Because reproduce at all indicate that the advantage of a winter these seeds were collected directly from the ®eld, these annual strategy and autumn germination in the tem- differences could be due to both genetic differences perate climate of Kentucky is due to the developmental between populations and to maternal effects accom- head start that enables plants to reproduce both earlier panying different ®eld conditions. The combined in- and at a larger size. ¯uences cause variation in important life-history char- The total failure of spring germinants to reproduce, acters and their plasticity even at a small regional scale and the altered development of those with an arti®cially in A. thaliana. protracted growing season, suggests that spring ger- The results of this experiment demonstrate a corre- mination by these Kentucky populations is not likely lation between germination timing and postgermination to be a successful life-history strategy. However, these life-history characters. They also show that postger- seeds experienced somewhat anomalous conditions mination life-history characters are under selection. from the beginning; all seeds were strati®ed (treated The combined results suggest that selection on ger- with cold as a seed) for only three days before being mination timing can operate indirectly through selec- put into the ®eld. Natural spring germinants in Ken- tion on postgermination characters. Thus selection on tucky would have been strati®ed much longer before later life-history stages can potentially in¯uence the germination. The duration of strati®cation and vernal- evolution of germination timing in a general manner. ization (cold treatment as a plant) has been shown to This experiment also demonstrated that the timing in¯uence the probability and timing of reproduction in of germination signi®cantly in¯uenced the selective en- Arabidopsis thaliana (Nordborg and Bergelson 1999). vironment experienced at later life stages. Evans and Therefore, the coordination between strati®cation and Cabin (1995) have argued that germination timing can vernalization was disrupted in these arti®cial spring strongly in¯uence the evolution of postgermination germinants, and neither the seed nor the rosette re- life-history characters (see also Tyler et al. 1978) and ceived an adequate duration of cold for normal devel- can thereby foster adaptive associations between ger- opment. Consequently, the extremely poor perfor- mination and postgermination characters. Others have mance of the spring germinants was probably due to a demonstrated theoretically that the timing of germi- combination of an inappropriate delay in germination nation can in¯uence the evolution of other characters which caused a fatal delay in reproduction, and a dis- (Venable and Lawlor 1980, Ritland 1983, Brown and rupted developmental program caused by the unusual Venable 1986, Venable and Brown 1988), and thereby experimental conditions. This experiment therefore cause associations between germination and nonger- demonstrated a distinct constraint on plasticity in the mination characters, but there exist few empirical stud- development of important life-history characters that ies to test the selective mechanisms for such associa- is very likely caused by absolute requirements for a tions. Many such studies focus on the selective con- minimal period of cold, either as a seed or a plant. In sequences of dormancy as a risk-reducing strategy in addition, it suggests an adaptive coordination of strati- a variable environment (e.g., (Venable and Lawlor ®cation and vernalization requirements which is medi- 1980, Brown and Venable 1986, Klinkhammer et al. ated by germination timing. Germination timing in¯u- 1987, Venable and Brown 1988), and they investigate ences the duration of strati®cation, and short strati®ca- the coevolution of other risk-reducing strategies such tion duration may be associated with natural selection as seed dispersal. The studies by Evans and Cabin that favors longer vernalization requirements to avoid (1995) suggest a broader in¯uence of germination tim- ¯owering during the winter. Clearly, the unnatural ex- ing on character evolution beyond its role as a risk- perimental combination of short strati®cation and short reducing trait; they demonstrate how germination in vernalization duration was associated with lower ®tness response to particular environmental conditions can de- in this sample than the more natural combination of termine the selective environment in which the ger- April 2002 GERMINATION TIMING INFLUENCES SELECTION 1015 minated seedlings and plants grow. In their system, 1999) and selective root placement (e.g., Birch and Lesquerella fendleri, which experiences variable mois- Hutchings 1994) can result in altered nutrient or water ture conditions that strongly in¯uence reproduction, conditions for plants. Habitat selection is probably both germination percentage and postgermination traits more common in plants than has previously been ap- vary with moisture conditions. The conditions under preciated. How such behaviors alter subsequent selec- which seeds germinate can potentially mediate the tion on other characters has not been well explored. moisture-dependent selection on postgermination char- In plants, habitat selection is frequently affected by acters. Supporting their hypothesis that germination be- the maternal parent. Maternal parents can in¯uence the havior can in¯uence the evolution of postgermination selective environment experienced by their progeny by characters, this study of Arabidopsis showed a direct altering dispersal (Donohue 1999) or requirements for relationship between the timing of germination and se- germination (reviewed in Gutterman 1992, and Baskin lection on postgermination life-history characters. and Baskin 1998; see also Kugler 1951, Dobrovolna and Because germination timing is often environmentally Cetl 1966, Gutterman 1978, Goto 1982, Roach and labile, this result also has implications for the evolution Wulff 1987, Biere 1991a, b, Platenkamp and Shaw 1993, of life-history characters under conditions of environ- LeÂon-Kloosterziel et al. 1994). This critical consequence mental change. For example, if autumns become drier of maternal effects can strongly in¯uence not only the or warmer, a postponement of germination timing until evolution of postgermination life-history characters, as later in the autumn could cause changes in natural se- discussed above, but it can in¯uence the adaptive value lection on life-history characters similar to the differ- of the maternal effect itself. This is because the adaptive ences in selection observed in this study. value of maternal effects depends on the degree of co- It is interesting to note that such changes in selection variance between maternal and progeny selective en- on life-history characters can occur even if the envi- vironments and on how well maternal parents predict ronment experienced by adult plants is not altered. In the selective environments of their progeny and prepare this experiment, all plants experienced the same en- their progeny for them (Donohue and Schmitt 1998). vironmental conditions in the spring, yet the different In summary, germination timing in this winter an- treatments were at different developmental stages. nual strongly in¯uenced not only the expression of life- Thus, germination timing can in¯uence natural selec- history characters but selection on those characters as tion on postgermination characters through two mech- well. By altering germination timing, plants can alter anisms. First, it can alter the selective environment their selective environment and the developmental experienced by plants. In this experiment, germination stage that is exposed to selection. This ability has im- timing determined whether or when seedlings would portant consequences for life-history evolution and for be exposed to winter conditions. However, this differ- the evolution of plasticity and maternal effects on ger- ence in environmental conditions had little effect on mination. These results also demonstrate that alter- ®tness in this study. Second, it can alter the develop- ations in germination timing that may accompany en- mental stage of the plant that is exposed to particular vironmental and evolutionary changes are expected to selective environments. It was this latter mechanism have important consequences for the expression and that caused the most difference in natural selection on evolution of postgermination life-history characters. life-history characters in this experiment. Thus stage- ACKNOWLEDGMENTS speci®c natural selection is likely to be in¯uenced signi®cantly by germination timing in plant populations. The author is grateful to Converse Grif®th and Christine Soliva for assistance with ®eldwork and to Lisa Dorn, Jo- The ability of plants to in¯uence their own selective hanna Schmitt, Carol Baskin, Jerry Baskin, and Converse environment has important consequences for the evo- Grif®th for many inspiring discussions on seasonal dormancy lution of life histories. Habitat selection in other or- and for comments on the manuscript. Susan Kalisz and an ganisms has been shown theoretically (e.g., Jones and anonymous reviewer gave several helpful suggestions for im- Probert 1980, Templeton and Rothman 1982, Rausher proving the manuscript. Carol and Jerry Baskin kindly al- lowed me to collect seeds from their garden. This research 1984) and empirically (Jaenike and Holt 1991, Barker was funded by a University of Kentucky Summer Faculty 1992) to in¯uence adaptive evolution and contribute to Research Fellowship, by the T. H. Morgan School of Bio- the maintenance of genetic variation. Habitat selection logical Sciences and the College of Arts and Sciences at the in plants has received comparatively less attention be- University of Kentucky, and by NSF grant #DEB0079489. cause of their sessile habit, but plasticity such as ger- LITERATURE CITED mination cueing, seed dispersal, or morphological and Alpert, P. 1999. Effects of clonal integration on plant plas- physiological modi®cations in response to the local en- ticity in Fragaria chiloensis. Plant Ecology 141:99±106. vironment all have the potential to alter the environ- BallareÂ, C. L., A. L. Scopel, and R. A. Sanchez. 1990. Far- ment experienced by plants. For example, shade-avoid- red radiation re¯ected from adjacent leaves: an early signal ance responses (Givnish 1982, BallareÂ et al. 1990) and of competition in plant canopies. Science 247:329±332. Barker, J. S. F. 1992. Genetic variation in cactophilic Dro- phototropism (Kendrick and Kronenberg 1994) cause sophila for oviposition on natural yeast substrates. Evo- plants to occupy altered light environments. Clonal for- lution 46:1070±1083. aging (e.g., MacDonald and Lieffers 1993, Alpert Baskin, C. C., and J. M. Baskin. 1998. Seeds: ecology, bio- 1016 KATHLEEN DONOHUE Ecology, Vol. 83, No. 4

geography and evolution of dormancy and germination. in Campanula americana (Campanulaceae): population dif- Academic Press, San Diego, California, USA. ferentiation. American Journal of Botany 81:521±527. Baskin, J. M., and C. C. Baskin. 1983. Seasonal changes in Kendrick, R. E., and G. H. M. Kronenberg. 1994. Photo- the germination responses of buried seeds of Arabidopsis morphogenesis in plants. Kluwer Academic, Dordrecht, thaliana and ecological interpretation. Botanical Gazette The Netherlands. 144:540±543. Klinkhammer, P. G. L., T. J. DeJong, A. J. Metz, and J. Val. Biere, A. 1991a. Parental effects in Lychnis ¯os cuculi: seed 1987. Life history tactics of annual organisms: the joint size, germination and seedling performance in a controlled effect of dispersal and delayed germination. Theoretical environment. Journal of Evolutionary Biology 4:447±465. Population Biology 32:127±156. Biere, A. 1991b. Parental effects in Lychnis ¯os cuculi: se- Kugler, I. 1951. Untersuchungen uber das Keimverhalten ei- lection on time of emergence and seedling performance in niger Rassan van Arabidopsis thaliana (L.) Heynh. beitrag the ®eld. Journal of Evolutionary Biology 3:467±486. zur Atiologie der Blutenbildung. Beitraege zur Biologie der Birch, C. P. D., and M. J. Hutchings. 1994. Exploitation of P¯anzen 28:173±210. patchily distributed soil resources by the clonal herb Gle- Lande, R., and S. J. Arnold. 1983. The measurement of se- choma hederacea. Journal of Ecology 82:653±664. lection on correlated characters. Evolution 37:1210±1226. Brown, J. S., and D. L. Venable. 1986. Evolutionary ecology Leon-Kloosterzeil, K. M., C. J. Keijzer, and M. Koorneef. of seed-bank annuals in temporally varying environments. 1994. A seed shape mutant Arabidopsis that is affected in American Naturalist 127:31±47. integument development. Plant Cell 6:385±392. Chouard, P.1960. Vernalization and its relations to dormancy. MacDonald, S. E., and V. J. Lieffers. 1993. Rhizome plas- Annual Review of Plant Physiology 11:191±237. ticity and clonal foraging of Calamagrostis canadensis in Dobrovolna, J., and I. Cetl. 1966. An increase of germination response to habitat heterogeneity. Journal of Ecology 81: of dormant seeds by pricking. Arabidopsis Information Ser- 769±776. vice 3. [Available online.] Masuda, M., and I. Washitani. 1992. Differentiation of spring Donohue, K. 1999. Seed dispersal as a maternally-in¯uenced emerging and autumn emerging ecotypes in Galium spur- character: mechanistic basis of maternal effects and selec- ium var. echinospermon. Oecologia 89:42±46. tion on maternal characters in an annual plant. American Napp-Zinn, K. 1976. Population genetical and geographical Naturalist 154:674±689. aspects germination and ¯owering in Arabidopsis thaliana. Donohue, K., D. Messiqua, E. Hammond-Pyle, S. M. Heschel, Arabidopsis Information Service 13. [Available online.] and J. Schmitt. 2000. Evidence of adaptive divergence in Nordborg, M., and J. Bergelson. 1999. The effect of seed plasticity: density- and site-dependent selection on shade and rosette cold treatment on germination and ¯owering avoidance responses in Impatiens capensis. Evolution 54: time in some Arabidopsis thaliana (Brassicaceae) ecotypes. 1956±1968. American Journal of Botany 86:470±475. Donohue, K., and J. Schmitt. 1998. Maternal environmental Platenkamp, G. A. J., and R. G. Shaw. 1993. Environmental effects: adaptive plasticity? Pages 137±158 in T. A. Mous- and genetic maternal effects on seed characters in Neo- seau and C. W. Fox, editors. Maternal effects as adapta- mophila menziesii. Evolution 47:540±555. tions. Oxford University Press, Oxford, UK. Ratcliffe, D. 1965. Germination characteristics and their in- Effmertova, E. 1967. The behaviours of ``summer annual,'' ter- and intra-population variability in Arabidopsis. Ara- ``mixed,'' and ``winter annual'' natural populations as com- bidopsis Information Service 13. [Available online.] pared with early and late races in ®eld conditions. Arabi- Rausher, M. D. 1984. The evolution of habitat preference in dopsis Information Service 4. [Available online.] a subdivided population. Evolution 38:596±608. Evans, A. S., and R. J. Cabin. 1995. Can dormancy affect Ritland, K. 1983. The joint evolution of seed dormancy and the evolution of post-germination traits? The case of Les- ¯owering time in annual plants living in variable environ- querella fendleri. Ecology 76:344±356. ments. Theoretical Population Biology 24:213±243. Evans, J., and D. Ratcliffe. 1972. Variation in ``after-ripen- Roach, D. A., and R. D. Wulff. 1987. Maternal effects in ing'' of seeds of Arabidopsis thaliana and its ecological plants. Annual Review of Ecology and Systematics 18: signi®cance. Arabidopsis Information Service 9:3±5. 209±235. Givnish, T. J. 1982. On the adaptive signi®cance of leaf SAS Institute. 1990. SAS/STAT User's Guide. SAS Institute, height in forest herbs. The American Naturalist 120:353± Cary, North Carolina, USA. 381. Templeton, A. R., and E. D. Rothman. 1982. Evolution in ®ne- Goto, N. 1982. The relationship between characteristics of grained environments: habitat selection as a homeostatic seed coat and dark-germination by gibberellins. Arabidop- mechanism. Theoretical Population Biology 19:326±340. sis Information Service 19:29±38. Tyler, B., M. Borrill, and K. Chorlton. 1978. Studies in Fes- Gutterman, Y. 1978. Seed coat permeability as a function of tuca. Observations on germination and seedling cold tol- photoperiodical treatments of the mother plants during seed erance in diploid Festuca pratensis and tetraploid F. pra- maturation in the desert annual plant: Trigonella arabica, tensis var. apennina in relation to their altitudinal distri- Del. Journal of Arid Environments 1:141±144. bution. Journal of Applied Ecology 15:219±226. Gutterman, Y. 1992. Maternal effects on seeds during de- Venable, D. L. 1984. Using intraspeci®c variation to study velopment. Pages 27±59 in M. Fenner, editor. Seeds: the the ecological signi®cance and evolution of plant life-his- ecology of regeneration in plant communities. Redwood tories. Pages 166±187 in R. Dirzo and J. Sarukhan, editors. Press, Melksham, UK. Perspectives in plant population ecology. Sinauer Associ- Jaenike, J., and R. D. Holt. 1991. Genetic variation for habitat ates, Sunderland, Massachusetts, USA. preference: evidence and explanations. American Natural- Venable, D. L., and J. S. Brown. 1988. The selective inter- ist 137S:67±S90. actions of dispersal, dormancy, and seed size as adaptations Jones, J. S., and R. F. Probert. 1980. Habitat selection main- for reducing risk in variable environments. American Nat- tains a deleterious allele in a heterogeneous environment. uralist 131:360±384. Nature 287:632±633. Venable, D. L., and L. R. Lawlor. 1980. Delayed germination Kalisz, S. 1986. Variable selection on the timing of germi- and dispersal in desert annuals: escape in space and time. nation in Collinsia verna (Scrophulariaceae). Evolution 40: Oecologia 46:272±282. 479±491. Wade, M. J., and S. Kalisz. 1990. The causes of natural Kalisz, S., and G. M. Wardle. 1994. Life history variation selection. Evolution 44:1947±1955.